Qorix8 - Real work agents

Qorix8 – Plug-and-play AI agents that autonomously handle your tasks like marketing, emails, and scheduling. No setup, no prompts—just tell it what you need.

Comment

Inspiration

We were tired of AI “assistants” that still make you click, configure, and babysit. Our goal was a hands-free, out-of-the-box AI workforce that can read your intent once and quietly run the entire show—marketing campaigns, email follow-ups, calendar juggling, research, content generation—while you focus on strategy or simply log off.

What it does

Qorix8 spins up a team of specialized agents the moment a user types a goal (“Launch a product-hunt campaign next week” or “Keep my inbox at zero”). The agents:

  1. Interpret the goal → auto-design a multi-step workflow (DAG).
  2. Coordinate with each other (search, write, schedule, post, track metrics).
  3. Report progress in a human-readable timeline you can pause, tweak, or resume. No deployment, prompt-crafting, or API wiring needed—just intent in, results out.

How we built it

  • Frontend / UX: Next.js 14 + Tailwind; one-field “command bar” UI plus live status feed via WebSockets.
  • Agent runtime: Python micro-services orchestrated with LangGraph-style DAGs; each node wraps an LLM (GPT-4o-mini) plus tool adapters (SMTP, Google Calendar, Notion, HubSpot, etc.).
  • Memory & data: Postgres/pgvector for long-term context, Redis for short-term scratchpad.
  • Auth & billing: Supabase for auth, Stripe metered usage.
  • DevOps: Docker-compose for local dev, Fly.io for edge deployment; CI/CD with GitHub Actions. We wrote the whole stack in 7 days for the hackathon and open-sourced core agent templates.

Challenges we ran into

  • Self-planning reliability: Getting the planner agent to generate safe, executable workflows without loops or dead ends.
  • Tool chaos: Normalizing dozens of third-party APIs behind a common action schema.
  • Guardrails: Preventing runaway email sequences or calendar spam; we built a reversible “dry-run” layer and human-in-the-loop checkpoints.
  • Latency vs. transparency: Balancing real-time feedback with batch execution so users feel in control but never overwhelmed.

Accomplishments we’re proud of

  • A genuine zero-setup demo that schedules meetings and writes follow-up emails from a single sentence.
  • Dynamic DAG visualizer that updates live as agents spawn new subtasks.
  • Plugin SDK that let teammates add a “TweetStorm” agent in under an hour.
  • Passed 150 integration tests covering error recovery and retry logic.

What we learned

  • Users judge trustworthiness by clarity of progress, not model size.
  • Domain-specific agents (email, calendar) outperform one giant “do-anything” agent in both speed and quality.
  • Fast rollback + audit trails are essential for enterprise adoption.
  • Small, well-scoped prompts with structured tool calls beat verbose “mega-prompts”.

What’s next for Qorix8 – Real Work Agents

  1. Public beta with a marketplace of ready-made agents (SalesOps, Recruiting, Finance).
  2. Enterprise on-prem edition with SOC 2 guardrails.
  3. Mobile command bar + Slack integration for “intent anywhere”.
  4. Open agent-builder studio so power users can publish and monetize their own workflows.
  5. Research into agent-to-agent negotiation so teams of Qorix8 instances can coordinate across companies.

Qorix8: From intent to done—no knobs, no code, just results.

Built With

  • bolt.new
  • claude
  • cursor
  • pocketflow
  • python
  • react
  • supabase
  • typescripts
Share this project:

Updates